Multi-cabin liquid cargo pump remote intelligent joint control system for liquefied gas carrier

Abstract

The present application relates to the technical field of automatic control of liquid cargo delivery of a ship, in particular to a multi-cabin liquid cargo pump remote intelligent joint control system for a liquefied gas carrier. The system acquires liquid levels, cabin pressures, rotation speeds, driving currents, outlet pressures, running states, valve states and common pipeline pressures of each liquid cargo pump according to a unified time reference, constructs a dynamic hydraulic coupling graph according to gain relationships and propagation time delays between state variables, establishes a multi-pump finite time domain prediction state, decomposes liquid cargo pump control variables into common mode control components for maintaining total delivery targets and differential mode control components for correcting state deviations among cabins, configures phases according to coupling path propagation time delays, generates cooperative control instructions after feasibility projection and marginal contribution redistribution. The system can reduce common pipeline pressure superposition and reciprocating fluctuation caused by multi-pump synchronous regulation, maintain stable total delivery state, and coordinate unloading progress of each liquid cargo cabin.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for ship liquid cargo transportation, specifically a remote intelligent control system for multi-compartment liquid cargo pumps on liquefied gas carriers. Background Technology

[0002] During the loading and unloading of liquefied gas carriers, multiple independent cargo tanks are typically set up, each equipped with a corresponding cargo pump. These pumps are connected to a common delivery pipeline via branch pipelines, and remote start / stop and speed regulation are implemented by a centralized control module. Existing multi-tank cargo pump control systems typically generate control targets based on pre-set total delivery volume, common pipeline pressure, or single-tank level drop rate, which are then executed by the local control loops of each cargo pump. The local control loops calculate deviations based on the pump's outlet pressure, drive current, speed, and the corresponding cargo tank level, and adjust the pump speed using a fixed proportional relationship. When a cargo tank reaches a low level limit or the corresponding cargo pump experiences an abnormal load, the centralized control module limits the speed of that pump or shuts it down, while simultaneously allocating the remaining delivery tasks to other cargo pumps. While this approach has a clear structure and easy-to-deploy control logic, its fundamental assumption is that the cargo pumps are independent of each other, or that the impact of the common pipeline on each pump is represented only by a fixed correction coefficient. It fails to continuously describe the dynamic interrelationships formed by different cargo pumps through the shared pipeline.

[0003] In multi-pump parallel operation scenarios, changes in the rotational speed of one liquid cargo pump alter the flow rate and outlet pressure of the corresponding branch, causing a redistribution of the main pipeline pressure via the common delivery pipeline, and shifting the operating points of other liquid cargo pumps. Existing technologies typically employ master-slave control, sequential speed regulation, or uniform slope control to mitigate such mutual influences. Master-slave control selects one liquid cargo pump to maintain the common pipeline pressure, while the remaining pumps adjust according to the target flow rate; sequential speed regulation adjusts the pump speed one by one according to a preset priority, preventing multiple liquid cargo pumps from operating simultaneously; uniform slope control limits the amount of rotational speed change per unit time for each pump, keeping the pressure change in the common pipeline gradual. These methods usually rely on fixed pump performance curves, pipeline resistance parameters, and adjustment sequences. When the number of participating liquid cargo pumps, valve connectivity, liquid level, or delivery back pressure changes, the direction, intensity, and pressure response arrival time of the influence between pumps all change accordingly, making it difficult to maintain consistency between the original master-slave relationship and fixed adjustment sequence with the current hydraulic state.

[0004] To improve multivariate control capabilities, some existing solutions employ model predictive control (MMC). This model integrates the speeds of each cargo pump, the pressure in the common pipeline, the liquid levels in each compartment, and the target total throughput into a predictive model. Within a finite prediction time domain, it calculates pump speed combinations that satisfy the constraints. Such solutions typically establish a fixed coupling matrix based on the pipeline topology, equivalent resistance, and pump characteristics determined during the design phase. A weighted objective function is used to balance total throughput, liquid level balance, and control smoothing. When the predicted results exceed liquid level, pressure, or drive current limits, the corresponding pump speeds are then trimmed or re-solved. While this control method can handle multiple control objectives, the fixed coupling matrix struggles to reflect the time-varying hydraulic relationships formed after normal speed regulation, pump start-up and shutdown, and valve switching. Each pump's regulation is still directly solved as a single set of variables, lacking separation in the control direction between the total throughput target and the inter-compartment balance target. Furthermore, the predictive model typically only considers the response amplitude and does not coordinate the propagation delays of different influence paths, meaning multiple pump speed commands may still result in in-phase pressure responses in the common pipeline.

[0005] Therefore, the hydraulic coupling relationship formed by multiple cargo pumps via a common transport pipeline will shift with the operating pump combination, valve connection status, liquid level changes, and pipeline pressure range. Existing controllers cannot continuously obtain the direction, gain, and propagation delay of each influencing path based on actual operational responses, nor can they separate the adjustment amount for maintaining the overall transport target from the adjustment amount for correcting inter-tank state deviations and coordinate arrival times accordingly. When multiple local control loops or centralized predictive controllers synchronously adjust pump speeds based on their respective deviations, pressure disturbances generated by different cargo pumps may be superimposed in phase within the common transport pipeline, causing the operating points of other cargo pumps to shift again. The controller continues to correct in the opposite direction, forming a reciprocating adjustment. This closed-loop coupling process causes periodic pressure fluctuations in the common transport pipeline and mismatches in the unloading progress of each tank, making it difficult to stably coordinate multiple tank cargo pumps while maintaining the target overall transport state. Summary of the Invention

[0006] The purpose of this invention is to provide a remote intelligent control system for multi-compartment liquid cargo pumps on liquefied gas carriers, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The remote intelligent control system for multi-compartment liquid cargo pumps on liquefied gas carriers includes a status acquisition interface, a remote control processor, and a pump control execution interface. The status acquisition interface acquires the liquid level and pressure of each liquid cargo tank, as well as the speed, drive current, outlet pressure, operating status, associated valve status, and common pipeline pressure of each liquid cargo pump according to a unified time base. The remote control processor constructs a dynamic hydraulic coupling spectrum that updates with the operating conditions based on the time delay response relationship and gain correlation relationship between various state variables. It establishes a multi-pump finite time domain prediction state based on the dynamic hydraulic coupling spectrum and decomposes the liquid cargo pump control quantity into a common mode control component to maintain the overall delivery target and a differential mode control component to correct the state deviation between compartments. After performing phase configuration on the differential mode control component according to the propagation delay of the coupling path, it generates a coordinated control command for each liquid cargo pump. The pump control execution interface sends the collaborative control commands to the remote control loops of the corresponding liquid cargo pumps.

[0008] Preferably, the remote control processor sets the liquid cargo tank, liquid cargo pump and pipeline section with shared transportation relationship as map nodes respectively, and extracts the candidate time delay sequence between any liquid cargo pump speed change, start-stop change or associated valve status change and other map node status responses from the continuous control cycle. Perform directional consistency screening and response attenuation screening on the candidate delay sequences to obtain coupling edges that characterize the influence direction, influence gain, and propagation delay; The coupling edges are grouped and stored according to the current set of liquid cargo pumps, valve connection status, and pressure range of the common transport pipeline. The weights of the coupling edges are updated using the response residuals of adjacent control cycles to form a dynamic hydraulic coupling map corresponding to the current transport topology and operating conditions.

[0009] Preferably, the remote control processor constructs a multi-pump predicted state vector based on the target remaining unloading time, liquid level drop rate, tank pressure change, and outlet pressure and drive current of each liquid cargo tank, and calculates the mapping relationship between the control quantity of each liquid cargo pump and the overall transport state and inter-tank difference state based on the coupling gain matrix corresponding to the dynamic hydraulic coupling spectrum. The control direction that maintains the target total delivery volume or the target common delivery pipeline pressure is defined as the common mode control subspace, and the control direction that does not change the target total delivery volume and is used to adjust the differences in the predicted state of different liquid cargo tanks is defined as the differential mode control subspace. The common mode control component and the differential mode control component are obtained by orthogonal projection.

[0010] Preferably, the remote control processor establishes a sequence of expected response arrival times for each differential control component based on the propagation delay from each liquid cargo pump to the common transport pipeline node in the dynamic hydraulic coupling diagram, and performs staggered arrangement of the same-direction pressure response and overlapping arrangement of the opposite-direction pressure response according to the expected response arrival times. Within any control cycle, the arranged differential control components and the common control components are combined to form a candidate control vector. The candidate control vector is then subjected to a feasibility projection based on the continuity of the liquid cargo pump speed, the rate of change of the outlet pressure, and the allowable pressure range of the common delivery pipeline to obtain the coordinated control command corresponding to each liquid cargo pump.

[0011] Preferably, the remote control processor extracts control change segments from the normal speed regulation, planned start-up and shutdown, and associated valve switching process of the liquid cargo pump, and uses the stable state before the control change as the reference state to perform reference elimination and time alignment on the state increment of each graph node after the control change segment; The state increments that maintain the same response direction across multiple control change segments and whose propagation delay falls into the corresponding delay cluster are used to update the coupling edge weights. State increments that appear only in a single control change segment or are inconsistent with the pressure change direction of the common delivery pipeline are removed from the edge weight update data. When the gain change rate of the same coupling side remains in the same direction in the continuous control change segment and the propagation delay migrates from the original delay cluster to another delay cluster, a new operating condition group is established and the coupling side parameters of the original operating condition group are retained.

[0012] Preferably, the remote control processor calculates the total transport state trajectory corresponding to the common mode control component and the inter-tank deviation state trajectory corresponding to the differential mode control component within a finite prediction time domain, and converts the minimum liquid level constraint, liquid level drop rate constraint, drive current constraint, outlet pressure constraint and control quantity change rate constraint of each liquid cargo tank into a predictable state feasible domain. First, a common-mode reference solution that satisfies the total transport target is obtained within the feasible region of the predicted state. Then, a differential-mode correction solution is obtained in the null space that keeps the total transport trajectory corresponding to the common-mode reference solution unchanged. The candidate speed adjustment amount of each liquid cargo pump is formed according to the sum of the common-mode reference solution and the differential-mode correction solution.

[0013] Preferably, the remote control processor maps the expected response arrival time corresponding to each differential mode control component to a continuous scheduling time window, and divides the differential mode control components into in-phase superposition group, out-of-phase cancellation group and weak coupling group according to the influence direction and influence gain of the coupling edge. A mutually exclusive arrival time window is set for the differential mode control components in the in-phase superposition group, an overlapping arrival time window is set for the differential mode control components in the out-of-phase cancellation group, and the original control period is retained for the differential mode control components in the weakly coupled group. When any control component exceeds the speed change rate constraint of the corresponding liquid cargo pump after adjustment by the time window, the excess part will be deferred to the adjacent time window and the subsequent prediction state will be corrected synchronously.

[0014] Preferably, the remote control processor calculates the marginal contribution vector of each liquid cargo pump unit control increment to the peak pressure of the common transport pipeline, the deviation of the target total transport volume, and the deviation of the predicted unloading time between compartments, based on the dynamic hydraulic coupling spectrum, for the candidate control components that are reduced in the feasibility projection. A redistribution priority sequence is constructed based on the remaining liquid level, outlet pressure margin, drive current margin, and marginal contribution vector of the corresponding coupling edge of each liquid pump. Under the condition of keeping the pressure prediction trajectory of the common delivery pipeline within the allowable range, the reduced delivery control quantity is gradually distributed to other liquid pumps in the priority sequence to form a cooperative control command constrained by marginal contribution.

[0015] Preferably, after establishing a new operating condition group, the remote control processor encodes the set of pumps, valve connection combinations, pressure range of common delivery pipeline, and liquid level range participating in the current control cycle as operating condition indexes, and selects the coupling edge parameter with the smallest distance from the operating condition index from the original operating condition group as the initial parameter. Before the consistency verification of the response samples that simultaneously contain the liquid cargo pump speed change segment and the associated valve switching segment is completed, a recursive confidence weight is set for the initial parameters, and the newly collected response is fused with the original operating condition prediction response according to the recursive confidence weight. After the consistency verification is completed, only the coupling edges that have passed the consistency verification of direction and propagation delay are updated, while the coupling edges that have not obtained valid response samples are frozen, forming a dynamic hydraulic coupling map constrained by confidence migration.

[0016] Preferably, when the remote control processor performs the redistribution of the delivery control quantity, it divides the liquid cargo pumps in the priority sequence into a main receiving set and a compensation receiving set according to the predicted suction safety margin and the predicted remaining unloading time. First, find the first redistribution vector within the main receiving set that minimizes the peak pressure increment of the common delivery pipeline. Then, project the delivery control quantities not received by the first redistribution vector onto the feasible control subspace of the compensation receiving set to obtain the second redistribution vector. Propagation delay correction is performed on the first redistribution vector and the second redistribution vector respectively, and the order of the main receiving set and the compensation receiving set is adjusted in subsequent control cycles according to the deviation between the actual pressure response and the predicted pressure response.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Based on the liquid level, tank pressure, pump speed, drive current, outlet pressure, operating status, valve status, and common pipeline pressure under a unified time reference, the direction, gain, and propagation delay between control changes and node responses are extracted to establish a dynamic hydraulic coupling map that updates with the current set of operating pumps, valve connection status, and pipeline pressure range. Based on the coupling gain matrix formed by the map, the control quantities of each cargo pump are projected as common-mode control components to maintain the target total delivery volume or target pipeline pressure, and differential-mode control components to correct for differences in inter-tank states such as the rate of liquid level decline and the estimated remaining unloading time. The differential-mode control components are then staggered or overlapped according to the propagation delay of the coupling path. Thus, the common delivery target and the inter-tank coordination target are solved in different control subspaces, avoiding the mutual cancellation of the two types of regulation quantities; the pressure responses of each pump are combined according to the expected arrival time, reducing the concentrated superposition of unidirectional disturbances within the common pipeline; the coupling weights are updated with the operating response, ensuring that the control relationship corresponds to the current hydraulic state, suppressing pressure fluctuations caused by synchronous regulation of multiple pumps, and maintaining coordinated unloading progress across multiple tanks.

[0018] 2. The coupling relationship is updated using control change segments generated during normal speed regulation, planned start-up and shutdown, and valve switching. The system performs benchmark elimination, time alignment, directional consistency screening, and propagation delay clustering on state increments. During condition transitions, original operating parameters are retained, and new operating condition groups are established to reduce the entry of occasional disturbances into the coupling edge weights. Within the finite prediction time domain, the system first obtains the common-mode benchmark solution in the feasible region formed by constraints on liquid level, drive current, outlet pressure, and control variable change rate. Then, it obtains the differential-mode corrected solution in the null space without changing the overall delivery trajectory, ensuring that the control variable synthesis process maintains constraint consistency. For control components reduced in the feasibility projection, the system redistributes them according to the marginal contribution of each pump to the pipeline pressure peak, total delivery deviation, and inter-chamber unloading time deviation, and determines the acceptance sequence based on the suction safety margin, outlet pressure margin, and drive current margin. This processing ensures that coupled model updates, control variable decomposition, time delay scheduling, and task redistribution are within the same control link, reducing discontinuities in joint control caused by model mutations, control out-of-bounds errors, and concentrated load on a single pump. Attached Figure Description

[0019] Figure 1 A flowchart of the overall control system for the remote intelligent control system of multi-compartment cargo pumps on an LPG carrier; Figure 2 A flowchart for the construction and working condition migration and updating of dynamic hydraulic coupling maps; Figure 3 The flowchart shows the multi-pump finite-time domain prediction and propagation delay scheduling process based on common-mode and differential-mode control decomposition. Figure 4 A flowchart for feasibility projection of candidate control quantities, redistribution of marginal contributions, and feedback correction. Figure 5This is a coupling gain identification diagram; Figure 6 Common-mode and differential-mode control decomposition diagram; Figure 7 Contribute graphs to targets in the finite time domain; Figure 8 Scheduling diagram for the expected response arrival time; Figure 9 This is a priority map for redistribution. Detailed Implementation

[0020] In one embodiment, please refer to Figure 1 The remote intelligent joint control system for multi-compartment cargo pumps on liquefied gas carriers includes a status acquisition interface, a remote joint control processor, and a pump control execution interface. The status acquisition interface acquires the liquid level and pressure of each cargo compartment, as well as the speed, drive current, outlet pressure, operating status, associated valve status, and common pipeline pressure of each cargo pump according to a unified time reference. The remote joint control processor constructs a dynamic hydraulic coupling spectrum that updates with the operating conditions based on the time delay response relationship and gain correlation between each state quantity. Based on the dynamic hydraulic coupling spectrum, it establishes a multi-pump finite-time domain predicted state and decomposes the cargo pump control quantity into a common-mode control component to maintain the overall delivery target and a differential-mode control component to correct the state deviation between compartments. After phase configuration of the differential-mode control component according to the propagation delay of the coupling path, it generates a coordinated control command for each cargo pump. The pump control execution interface sends the coordinated control command to the remote control loop of the corresponding cargo pump, thereby forming a closed-loop joint control link from operating status acquisition, coupling relationship identification, predicted state construction, control direction decomposition, propagation delay correction to control command execution.

[0021] In this embodiment, the status acquisition interface establishes a unified data frame for each cargo tank, each cargo pump, and the common delivery pipeline. Each status quantity in the data frame is accompanied by the acquisition time, source identifier, status category, and current validity flag. The remote control processor uses the unified clock provided by the ship control network as a reference to map data with different update cycles to the same control time. For data whose acquisition time is earlier than the current control time, a hold value is used and the hold duration is recorded. For data that changes between adjacent control times, piecewise linear interpolation is used to form an alignment value. For start-stop status and valve status, an event hold method is used to maintain the most recently determined state, so that the liquid level, tank pressure, speed, current, outlet pressure, and common delivery pipeline pressure are established on the same time axis, avoiding misjudging the state changes that do not occur at the same time as hydraulic coupling responses due to the inconsistent update times of different acquisition channels.

[0022] Specifically, the remote control processor treats each cargo tank as a cargo status node, the cargo pump corresponding to the cargo tank as a control node, and the branch merging section and common transport pipeline section with shared transport relationship as pipeline nodes. It determines whether there is a reachable path between nodes based on the status of associated valves. The cargo tank node stores the liquid level and tank pressure status, the cargo pump node stores the speed, drive current, outlet pressure and operating status, and the pipeline node stores the common transport pipeline pressure and the composite transport status derived from the upstream pump status. The directed edges between nodes are used to record the influence direction, influence gain, propagation delay and reliability weight when the state change of a certain control node is transmitted to another node. The path blocked by valve closure does not participate in the coupling calculation within the current operating condition group but retains historical parameters.

[0023] In a preferred embodiment, please refer to Figure 2 The remote control processor identifies naturally occurring control change segments from continuous control cycles. These control change segments include speed changes caused by normal speed regulation of the liquid cargo pump, operating status changes caused by planned start-stop, and path status changes caused by switching of associated valves. Before the change segment begins, the processor selects a continuously stable state interval as a reference interval and calculates the reference mean and natural fluctuation range of each node's state from the reference interval. Then, it extracts the state increment sequence of each node relative to the reference mean after the change segment and uses the starting time when the state increment exceeds the natural fluctuation range as the candidate response time. Therefore, it can form data samples that can be coupled and identified without applying additional experimental disturbances during the liquid cargo transportation process.

[0024] The initial influence gain of the coupling edge can be determined according to the following formula: ; in, R represents the gain of influence from control node a to response node b, and R represents the number of control change segments that pass the consistency screening. This represents the control increment of control node a within the r-th segment. This represents the state increment of the corresponding response node b after propagation delay correction. The sample weights are determined by fragment stability and response continuity. This indicates that to prevent positive numbers with a denominator of zero, the example value is that when the speed increments in the three segments are positive, positive, and negative respectively, and the pressure increments in the common delivery pipeline are positive, positive, and negative respectively, the product terms retain the same sign and form a positive result. Therefore, the coupling edge is determined to be a unidirectional influence edge, and a negative influence is formed when the direction of pressure increment is opposite to that of speed increment. And it was determined to be an edge with reverse influence.

[0025] Please refer to Figure 5 , Figure 5The horizontal axis represents the control change segment number, ranging from 1 to 95, while the vertical axis represents the standardized increment and weight. In the legend, the solid line represents the liquid pump control increment, the dashed dotted line represents the response increment, the hollow circle represents the sample weight, the dotted line represents the rolling gain, and the two horizontal reference lines are used to display different levels of engineering discrimination thresholds. Figure 5 This corresponds to the calculation process of the gain from the control node to the response node in the text, and is used to put the control changes, response changes and segment reliability in multiple natural control change segments into the same time series for observation.

[0026] from Figure 5 As can be seen, the sample weight curve remains between 0.64 and 0.91, indicating that the segments included in the calculation have high stability and no large-scale failure samples appear. The control increment curve shows alternating positive and negative values ​​throughout the entire sequence, with a peak value of approximately 0.76 and a trough value of approximately -0.75, providing sufficient directional identification information. The response increment lags behind the control increment by approximately 5 to 8 segments, and in multiple sections, it changes in the same direction as the control increment, indicating a stable transmission relationship between control changes and common pipelines or response nodes.

[0027] The rolling gain curve forms a predominantly positive interval around segments 30 to 44, and a negative interval around segments 8 to 22, with an overall amplitude ranging from approximately -0.60 to 0.72. This distribution indicates that the same control node may exhibit unidirectional or inverse effects under different connectivity states and pressure intervals. Therefore, a gain calculation with sample weights is used instead of directly overriding the coupling edge parameters with a single difference. Figure 5 This document provides a description that supports edge weight updates, direction selection, and reliability recursion in dynamic hydraulic coupling graphs.

[0028] The propagation delay is obtained by performing a correlation search within a limited time window on the control increment sequence and the response increment sequence. The remote control processor calculates the directional consistency rate and amplitude correlation for each candidate delay, and retains only the candidate delays whose directional consistency rate meets the preset sample majority condition and whose amplitude correlation reaches the statistical lower limit of the current operating condition. Then, adjacent candidate delays are aggregated into delay clusters and the propagation delay is calculated with the weighting center as the coupling edge. If a control node has two stable delay clusters for the same response node, a direct influence edge and an indirect influence edge formed through the shared pipeline are established respectively. In subsequent predictions, the response is accumulated according to the propagation delay and gain of the two edges to avoid compressing multi-path transmission into a single fixed delay.

[0029] Furthermore, the dynamic hydraulic coupling map forms a condition index based on the currently operating set of liquid cargo pumps, associated valve connection combinations, pressure range of common delivery pipelines, and liquid level range of each compartment. Coupled edges under the same condition index share a set of parameters, while parameters are stored independently between different condition indices. When the current condition index changes, the processor selects the parameter set from historical conditions that has the same set of operating pumps and valve connection combinations and whose pressure range is closest to the liquid level range as the initial migration value, and sets an initial confidence weight lower than that of the verified parameter set. This allows the controller to retain continuous control capability at the beginning of a new condition, while avoiding the direct acquisition of full control authority by historical parameters that have not been verified by the current response.

[0030] In this embodiment, the coupling edge weights are not directly overwritten based on a single deviation, but rather the current response and the original parameters are recursively merged according to their credibility. The corresponding credibility weight is increased only when the same coupling edge maintains the same influence direction in multiple consecutive control change segments and the propagation delay is in the same delay cluster. When the state increment only appears in a single segment, the response direction is inconsistent with the pressure change of the common pipeline, or the response occurs outside the physically reachable time window, the state increment does not enter the edge weight update. When the gain change rate remains in the same direction and the propagation delay stably migrates from the original delay cluster to another delay cluster, the processor establishes a new operating condition group and freezes the original operating condition group to distinguish between the real hydraulic relationship migration and short-term disturbances.

[0031] Please refer to Figure 3 In each control cycle, the remote control processor calls the dynamic hydraulic coupling graph based on the current operating condition index. It uses the speed control increment of each cargo pump as the control vector, and the liquid level drop rate of each tank, the expected remaining unloading time, the tank pressure change, the outlet pressure of each cargo pump, the drive current, and the pressure of the common delivery pipeline as the predicted state. It constructs a time-sharing coupling matrix based on the influence gain and propagation delay in the graph. Each column in the time-sharing coupling matrix corresponds to the control increment of a cargo pump at the current and subsequent control times, and each row corresponds to the change of a predicted state at the corresponding predicted time. The position where there is no reachable path in the matrix is ​​set to zero, and the position where there is multi-path coupling writes the gain into different predicted times according to the time delay of each path.

[0032] In a preferred embodiment, the estimated remaining unloading time is calculated based on the transportable cargo state quantity corresponding to the current liquid level and the smoothed liquid level drop rate. The liquid level drop rate uses a robust fitted value of the liquid level change in the most recent continuous control cycle instead of a single-point difference value. The processor removes transitional samples corresponding to valve switching and cargo pump start-up and shutdown. When the tank pressure change exceeds the stable range, the estimated remaining unloading time is marked as a state to be corrected. After the tank pressure returns to continuous change, the new estimate is written into the prediction state. This avoids liquid level measurement fluctuations and instantaneous operating condition switching from directly changing the differential mode control direction.

[0033] The common-mode control component and the differential-mode control component are decomposed according to the following formula: , , ; Where u represents the candidate control increment vector for all liquid cargo pumps. Indicates the common-mode control component. Let I represent the differential control components, and let I represent the identity matrix corresponding to the number of operating liquid cargo pumps. This represents the projection matrix formed by the overall transport state sensitivity vector. An example value is taken when two liquid cargo pumps have the same sensitivity to the overall transport state. The changes in the same direction and amount of the two pumps are mapped to common-mode control components, and the changes in the amount of the increase of one pump and the decrease of the other pump are mapped to differential-mode control components. When the two pumps have different sensitivities, the projection matrix allocates the control increment required for the same total delivery change according to the sensitivity ratio.

[0034] Please refer to Figure 6 , Figure 6 The horizontal axis represents the combination of the liquid cargo pump number and the prediction step size, covering the control sequences of P1 to P4 within multiple prediction steps; the vertical axis represents the speed adjustment amount. The legend represents the candidate quantity, common modulus, differential modulus, and composite quantity, respectively. Figure 6 The process described in the paper decomposes all candidate control vectors for liquid cargo pumps into common-mode control components and differential-mode control components. The common-mode component is responsible for the unified tracking of the overall transport status or common pipeline pressure, while the differential-mode component is responsible for the correction of the expected remaining time, liquid level drop rate, and tank pressure deviation of each compartment.

[0035] from Figure 6 It can be seen that the candidate quantity and the synthesized quantity curves basically overlap, indicating that the control sequence after decomposition and resynthesis maintains the original candidate control direction. The candidate quantity ranges from approximately -78 to 109 rpm, with the common modulus and candidate quantity dominating in terms of value, indicating that the main control requirement under the current operating condition comes from maintaining the overall delivery state. The differential modulus is relatively small, mainly distributed between -18 and 21 rpm, but its positive and negative directions switch between different pumps, reflecting the system's local correction effect on the differences in state between multiple compartments.

[0036] Figure 6The first two steps of P4 show significant positive adjustments, while the step size in P2 shows a significant negative adjustment, reflecting the differences in the coupling sensitivity of different cargo pumps in the common pipeline and the remaining unloading status of the compartment. This result indicates that the projection matrix in the paper is not simply distributed equally, but rather that the unidirectional changes are written into the common-mode subspace based on the total transport sensitivity, and the remaining changes that do not change the first-order prediction of the total transport are retained as differential-mode components, providing a clear control input structure for subsequent finite-time domain solutions and time-delay scheduling.

[0037] The overall transport state sensitivity vector is composed of the effective gains of each cargo pump to the common transport pipeline node in the dynamic hydraulic coupling spectrum, and is modified for availability based on the current outlet pressure, drive current and valve connection status. The processor uses the subspace spanned by the sensitivity vector as the common mode control subspace and the null space that keeps the first-order prediction of the overall transport state unchanged as the differential mode control subspace. The common mode control component is used to track the target total transport volume or the target common transport pipeline pressure, and the differential mode control component is used to adjust the deviation between the expected remaining unloading time of each compartment, the liquid level drop rate and the compartment pressure change, so that the two types of control objectives can be solved in different control directions.

[0038] Preferably, the remote control processor constructs a state trajectory within a finite prediction time domain. The current state serves as the prediction starting point, and the state at each subsequent prediction time is calculated jointly by the state at the previous prediction time, the corresponding control increment, and the time-sharing coupling matrix. The liquid level state is recursively deduced according to the changes in delivery. The outlet pressure and the pressure of the common delivery pipeline are written into the response according to the propagation delay of the coupling edge. The drive current is recursively deduced according to the joint response of the pump speed change and the outlet pressure change. The valve state remains unchanged within the prediction time domain according to the currently confirmed state. Only when a scheduled valve switching event is received will the reachable path be changed at the corresponding prediction time.

[0039] The control solution in the finite prediction time domain is obtained using the following formula: ; Where H represents the prediction time domain length, This represents the sequence of all common-mode control components in the prediction time domain. This represents the sequence of all differential control components in the prediction time domain. and These represent the overall transport status and its target value at the h-th prediction time, respectively. This represents the difference in state vectors between the estimated remaining unloading time and the rate of liquid level drop for each compartment. and These represent the pressure of the public transmission pipeline and its reference value, respectively. This indicates the incremental change in the liquid cargo pump control between adjacent prediction times. , , , These represent the non-negative weights of the four types of deviations. The relative order of these weights is determined by the operational objective rather than using fixed hardware parameters. For example, during the multi-compartment synchronous unloading phase, the weights are increased. Compared to The proportion should be increased when the pressure on public pipelines is limited. The proportion.

[0040] Please refer to Figure 7 , Figure 7 The horizontal axis represents the prediction step size, ranging from 1 to 8, and the vertical axis represents the contribution of the target item. The legend corresponds to the total transport item, inter-cabin item, pressure item, smoothing item, and total target, respectively. Figure 7 The objective function corresponding to the finite predictive time-domain control solution in the paper is used to observe the relative contributions of different deviation terms in multiple future control steps. The total transport term reflects the deviation between the total transport state and the target state, the inter-compartment term reflects the synchronicity of unloading in multiple compartments, the pressure term reflects the degree of deviation of the common pipeline pressure, and the smoothing term reflects the execution smoothness constraint caused by the change in adjacent control increments.

[0041] from Figure 7 As can be seen, the total transport term gradually decreased from approximately 0.92 in step 1 to approximately 0.34 in step 8, indicating that predictive control continuously reduced the total transport deviation during the time-domain progression. The inter-cabin term decreased from approximately 0.72 to approximately 0.18, indicating that the multi-cabin synchronization error was gradually weakened. The pressure term differed from the previous two types, being smaller in the early stage and rising to approximately 0.62 in the later stage, suggesting that as the transport target gradually approached, the common pipeline pressure became an important constraint for the subsequent verification.

[0042] The smoothing term reaches 1.00 in the first step and then rapidly decreases to 0.09. The overall objective also decreases from 1.00 to 0.20, indicating that the main cost of the current control sequence is concentrated in the initial adjustment phase, and the control changes in subsequent steps tend to be gradual. This trend is consistent with the processing logic of first obtaining the common-mode benchmark solution, then finding the differential-mode corrected solution in the null space, and performing feasible region verification. Figure 7 This indicates that the objective function does not solely pursue the minimum flow deviation, but rather involves a comprehensive trade-off between total transport, multi-compartment synchronization, pressure limits, and control smoothness.

[0043] In this embodiment, the solution process first obtains the common mode reference solution within the feasible region of the predicted state. The feasible region of the predicted state is jointly formed by the minimum liquid level boundary of each liquid cargo tank, the liquid level drop rate boundary, the drive current boundary of each liquid cargo pump, the outlet pressure boundary, the pressure boundary of the common transport pipeline, and the control increment change rate boundary. The boundary values ​​are read from the allowable operating range provided by the ship's existing control system and are not fixed in the joint control system. The processor only accepts the common mode control sequence that simultaneously satisfies the boundary conditions of all predicted times. Then, the differential mode correction solution is obtained in the null space that keeps the total transport trajectory corresponding to the common mode reference solution unchanged, so that the inter-tank difference state is corrected without changing the already determined total transport trajectory.

[0044] If the superposition of the differential mode correction solution and the common mode reference solution causes any predicted state to leave the feasible region, the processor performs a feasibility projection on the synthesized control vector. During the projection process, the angle between the control direction and the original synthesized control vector is kept to a minimum, and the control increment of the out-of-bounds liquid cargo pumps is compressed. If multiple liquid cargo pumps approach the boundary at the same time, the compression ratio is jointly determined according to the gain of each pump's influence on the pressure of the common delivery pipeline and the current remaining control margin. The projected control quantity is then re-sent into the prediction model for verification until the liquid level, pressure, drive current, and control rate of change in the prediction time domain are all within the allowable range.

[0045] In another embodiment, the remote control processor establishes an expected response arrival time for each differential control component based on the propagation delay from each liquid cargo pump to the common delivery pipeline node in the dynamic hydraulic coupling graph. The expected response arrival time is obtained by adding the planned sending time of the control command, the transmission delay of the pump control execution interface, the response delay of the local control loop of the liquid cargo pump, and the propagation delay of the coupling edge. The transmission delay and the local response delay are recursively estimated by the difference between the sending receipt of the most recent control command and the actual change time of the rotational speed. The propagation delay of the coupling edge is determined by the delay cluster formed by the aforementioned response samples.

[0046] The expected response arrival time can be calculated using the following formula: ; in, This indicates the estimated time when the pressure response of the i-th liquid cargo pump will reach the common pipeline. Indicates the planned time for sending coordinated control commands. This indicates the transmission delay of the pump control execution interface. This indicates the local response delay of the corresponding liquid cargo pump remote control loop. This represents the hydraulic propagation delay of the corresponding coupling path. An example value is given where the differential mode control components of both liquid cargo pumps cause a pressure increase in the common delivery pipeline, and both... When they fall into the same scheduling time window, the processor delays the control component with the larger marginal pressure contribution, causing the two expected arrival times to separate. When the pressure changes caused by the two differential control components are in opposite directions, the processor adjusts the two expected arrival times to be within the same time window.

[0047] Please refer to Figure 8 , Figure 8 The horizontal axis represents the liquid cargo pump number and control segment number, while the vertical axis represents relative time in seconds. The legend shows the planned delivery time, transmission delay, local response delay, hydraulic propagation delay, estimated arrival time, and the start and end points of the window. Figure 8The calculation process corresponding to the response arrival time in the text is used to unify the four types of time quantities—command sending, interface transmission, remote control loop response, and pipeline propagation—onto the same time axis.

[0048] from Figure 8 As can be seen, the planned transmission time for P1 is 12.0 seconds, and after adding transmission, local response, and hydraulic propagation, its expected arrival time is 32.0 seconds; the planned transmission times for P2 and P3 are both 42.0 seconds, but due to differences in hydraulic propagation and local response, their expected arrival times are approximately 65.0 seconds and 66.5 seconds, respectively; the planned transmission time for P4 is 72.0 seconds, and its expected arrival time is approximately 92.6 seconds. The arrival point of each pump falls between the start and end points of its corresponding window, indicating that the scheduling results satisfy the time window constraint.

[0049] Figure 8 It also shows that although P2 and P3 belong to similar sending batches, their arrival times are controlled within an identifiable adjacent interval, which helps to avoid the concentrated superposition of pressure responses in the same direction in the common pipeline; P4 is arranged in a later window to reduce the risk of pressure peak superposition in the later stage. Figure 8 The phase configuration proposed in the explanatory paper does not only change the order of command transmission, but also incorporates communication, execution and hydraulic propagation delays into the scheduling, so that differential mode control can achieve pressure response peak shifting without changing the common mode transport trajectory.

[0050] Specifically, the processor divides the differential mode control components into in-phase superposition group, out-of-phase cancellation group, and weak coupling group based on the influence direction and influence gain of the coupling side. The control components in the in-phase superposition group correspond to the pressure response of the common transmission pipeline in the same direction and are set with mutually exclusive arrival time windows. The control components in the out-of-phase cancellation group correspond to the pressure response in the opposite direction and are set with coincident arrival time windows. The influence of the control components in the weak coupling group on the pressure of the common transmission pipeline is lower than the statistical range of the current operating condition and the original control cycle is retained. The scheduling result only adjusts the transmission time and segmentation method of the control components, without changing the total transmission trajectory determined by the common mode control components.

[0051] When a differential mode control component exceeds the allowable rate of change of the corresponding liquid cargo pump after adjustment within a time window, the processor splits the control component into continuous sub-components. Sub-components that satisfy the rate of change boundary of the current control cycle are placed in the current time window, while the remaining sub-components are sequentially deferred to subsequent adjacent time windows. After each deferred period, the processor synchronously updates the outlet pressure, drive current, common pipeline pressure, and inter-cabin difference status within the finite prediction time domain. If the predicted status after the deferred period cannot be kept within the feasible domain, the differential mode correction amount is reduced while the common mode reference solution is retained.

[0052] In this embodiment, the pump control execution interface receives a coordinated control command synthesized from a common-mode control component and a differential-mode control component with phase configuration. It generates an independent command frame according to the liquid cargo pump identifier, the planned execution time, and the control increment, and sends the command frame to the remote control loop of the corresponding liquid cargo pump. The remote control loop returns the receiving time, the planned execution time, and the actual speed change status. The remote control processor uses the receipt to update the transmission delay and local response delay. If a certain command does not generate a corresponding speed change at the planned execution time, the unrealized control increment is withdrawn from the current prediction state and the subsequent control sequence is recalculated.

[0053] In a preferred embodiment, please refer to Figure 4 The remote control processor performs marginal contribution analysis on the candidate control components that are reduced in the feasibility projection. For each cargo pump that can handle the liquid, it applies a small virtual control increment that does not change the current operating condition index. It also calculates the changes in the peak pressure of the common transport pipeline, the deviation of the target total transport volume, and the deviation of the predicted unloading time between compartments through the time-sharing coupling matrix. The three types of changes form the marginal contribution vector of the corresponding cargo pump. The virtual control increment is only used for calculation within the model and is not sent to the pump control execution interface.

[0054] Marginal contribution and redistribution priority are calculated according to the following formula: ; in, This indicates the redistribution priority of the i-th liquid cargo pump. Indicates the predicted inhalation safety margin. Indicates the margin of export pressure. Indicates the drive current margin. This represents the marginal contribution of a unit control increment to the peak pressure of the public transport pipeline. This represents the marginal contribution of a unit control increment to the deviation from the target total transport volume. This represents the marginal contribution of a unit control increment to the predicted unloading time deviation between compartments. , , , , , Indicates the non-negative portfolio weights. This indicates a positive number to prevent the denominator from being zero. An example value is given when two candidate liquid cargo pumps have similar safety margins; the pump with the smaller marginal contribution to the peak pressure in the common delivery pipeline receives a larger value. And it will take priority in handling the reduced transport control volume.

[0055] Please refer to Figure 9 , Figure 9The horizontal axis represents the number of liquid cargo pumps that can be accepted, and the vertical axis represents the priority and normalized margin. The legend represents the priority, safety margin, pressure margin, current margin, marginal contribution, and normalized acceptance quantity, respectively. Figure 9 The calculation process corresponding to the marginal contribution and redistribution priority in the text is used to simultaneously compare the safety margin, outlet pressure margin, drive current margin, and the impact of unit control increment on common pipeline pressure, total delivery deviation, and inter-compartment differences for each cargo pump.

[0056] from Figure 9 As can be seen, P2 has the highest priority, approximately 3.40, and also the largest normalized acceptance amount, indicating that its combination of comprehensive safety margin and marginal impact is most suitable for accepting the compressed control quantity; P1 has a priority of approximately 2.28, placing it in the second acceptance sequence; although P3 has good current and pressure margins, its marginal contribution is high, indicating that further increasing its control quantity is more likely to amplify the target deviation or pressure peak, thus its priority is lower than P1 and P2; P4 has a high safety margin, but low pressure and current margins, resulting in the smallest final acceptance amount.

[0057] This result indicates that the redistribution is not based on an average allocation according to the number of liquid pumps or their rated capacity, but rather on a comprehensive evaluation that considers suction safety, outlet pressure, current constraints, and marginal effects within the same prioritization framework. When a pump's control input is reduced due to level, pressure, or current boundaries... Figure 9 The sorting shown can determine the main receiving set and the compensation receiving set, and provide the receiving pump sequence for subsequent propagation delay correction, supporting the maintenance of the feasible domain and the suppression of pressure peaks in the multi-compartment joint control process.

[0058] The processor constructs a receiving sequence based on the redistribution priority and divides the candidate cargo pumps into a main receiving set and a compensation receiving set according to the predicted suction safety margin and the predicted remaining unloading time. The main receiving set includes cargo pumps with a large safety margin and a predicted remaining unloading time that is not less than the current multi-compartment median state. The compensation receiving set includes cargo pumps that are still in the feasible region but have a small safety margin or remaining unloading time. The processor first finds the first redistribution vector in the main receiving set that minimizes the peak pressure increment of the common delivery pipeline, and then projects the unreceived delivery control quantities to the feasible control subspace of the compensation receiving set to form the second redistribution vector.

[0059] Both the first and second reassignment vectors need to be corrected for propagation delay. The processor rearranges the transmission time of the control increment based on the propagation delay of each receiving liquid cargo pump to the common transport pipeline node, so that the pressure response formed by the receiving control quantity does not arrive in the same direction as the original response. If there is a continuous deviation in the same direction between the actual pressure response and the predicted response after reassignment, the ranking of the corresponding receiving liquid cargo pump will be reduced and the marginal pressure contribution estimate will be increased in the next control cycle. If the actual response and the predicted response maintain the same direction and timing, the current ranking will be maintained and the response sample will be used to update the coupling edge weights.

[0060] Furthermore, when the current operating condition index is not completely consistent with the historical operating condition index and the new coupling edge parameters have not been verified by sufficient response samples, the processor uses a trusted migration method to generate a transient dynamic hydraulic coupling map. Historical parameters that are completely consistent with the pump set and valve connection combination participating in the operation receive higher migration weights, historical parameters that are adjacent to the pressure range or liquid level range of the common delivery pipeline receive lower migration weights, and historical parameters with different reachable paths do not participate in the migration. The newly acquired response and the historical predicted response are merged according to the migration weights, and only the coupling edges that have passed the consistency of direction and propagation delay are unfrozen.

[0061] In this embodiment, if the response sample containing both the liquid cargo pump speed change segment and the associated valve switching segment has not yet completed consistency verification, the initial parameters in the transient spectrum decrease the confidence weight according to the control cycle. As new response samples enter, the processor compares the response direction, propagation delay, and gain changes before and after the speed change and valve switching, respectively. Only when the valve connection relationship changes and the corresponding coupling edge shows an addition, attenuation, or disappearance consistent with the reachable path, and the liquid cargo pump speed change response and change mutually confirm each other, will the new operating condition grouping be confirmed and the corresponding coupling edge be converted to an updatable state.

[0062] In another preferred embodiment, when the system is in a multi-compartment synchronous unloading state, the remote control processor constructs the inter-compartment differential state based on the dispersion of the expected remaining unloading time of each compartment, and determines the differential mode correction direction based on the current liquid level drop rate and the available control margin of the pump. The liquid cargo pump corresponding to the liquid cargo compartment with a shorter expected remaining unloading time obtains a differential mode component that reduces the transport control amount, and the liquid cargo pump corresponding to the liquid cargo compartment with a longer expected remaining unloading time obtains a differential mode component that increases the transport control amount. After all differential mode components are projected into the common mode null space, the first-order prediction of the overall transport state remains unchanged, and the in-phase superposition of the pressure in the common transport pipeline is limited by the propagation delay scheduling.

[0063] When a cargo tank approaches the minimum liquid level boundary, the processor shrinks the feasible control subspace of the corresponding cargo pump in the direction of reducing the delivery control quantity, and assigns the common mode control quantity that can no longer be undertaken as the quantity to be redistributed to the marginal contribution redistribution mechanism. Other cargo pumps do not take on the control quantity in a fixed proportion, but take it on according to the priority sequence formed by the suction safety margin, the outlet pressure margin, the drive current margin, and the marginal contribution of the pressure peak. The control component after taking it on re-enters the phase configuration and finite time domain verification until the subsequent predicted state is back in the feasible domain.

[0064] When a planned shutdown occurs at a participating liquid cargo pump, the processor gradually decays the future control column of the corresponding liquid cargo pump from the time-sharing coupling matrix before executing the shutdown command. It then distributes the common-mode control quantity undertaken by the pump to the still-running liquid cargo pump and reprojects the differential-mode control quantity undertaken by the pump to a new differential-mode subspace. The pressure response caused by the actual speed drop of the shut-down pump is written into the predicted state according to the propagation delay of the coupling path. The compensation control responses of other liquid cargo pumps are arranged in different scheduling time windows before and after the arrival of the pressure response to avoid the shutdown disturbance and the compensation speed increase from being simultaneously superimposed.

[0065] When a planned switch occurs in an associated valve, the processor saves the current dynamic hydraulic coupling graph and predicted state before the switch event. At the predicted time corresponding to the switch event, it updates the node reachability relationship and enables the operating condition group corresponding to the new valve connection combination. If there is no verified operating condition group, it calls the confidence migration parameter. The first batch of control change segments after the valve switch is only used to verify the coupling edge direction and propagation delay. The gain is not immediately and significantly corrected. The edge weight is updated recursively after multiple response segments are consistent. This limits the short-term pressure changes caused by valve transients from entering the long-term coupling parameters.

[0066] Specifically, during system operation, each control cycle performs state alignment, operating condition index matching, dynamic graph calling, predicted state construction, common mode benchmark solving, differential mode correction solving, propagation delay scheduling, feasibility projection, marginal contribution redistribution, and cooperative control command issuance. After the command is executed, the prediction results are verified by changes in actual speed, outlet pressure, drive current, liquid level, and common delivery pipeline pressure. The verification results are used to update transmission delay, local response delay, coupling side propagation delay, and impact gain. Responses that fail the consistency verification are only retained as abnormal samples and are not written into the current operating condition parameters.

[0067] The remote control processor saves the operating condition index, dynamic graph version, common mode control components, differential mode control components, expected response arrival time, feasibility projection results, marginal contribution vector, redistribution sequence, and execution receipt used for each control solution. It also associates these data with actual state changes through the same time base. When the pressure of the common transmission pipeline deviates from the predicted trajectory, the processor traces back the corresponding control components and propagation paths along the coupling edges in the dynamic graph to determine the coupling edges that need to have their confidence weights reduced and the transmission delays or local response delays that need to be re-estimated, so that the model correction targets remain clear.

[0068] In this embodiment, all calculations of the remote control processor are completed before the control command of the liquid cargo pump is issued. The pump control execution interface only receives the control increment after finite prediction time domain verification, propagation delay configuration and feasibility projection. It does not directly receive the undecomposed total delivery deviation or inter-cabin difference deviation. The remote control loop of the liquid cargo pump continues to execute the existing local speed closed loop. The control system achieves multi-pump coordination by changing the target speed or target speed increment of each local loop without changing the structure of the liquid cargo pump body and the liquid cargo pipeline.

[0069] In one implementation, when the target total transport state is characterized by the pressure of the common transport pipeline, the processor constructs a common mode control target using the reference value of the common transport pipeline pressure and uses the total transport volume as the predicted constraint state. In another implementation, when the target total transport state is characterized by the sum of the estimated transport volumes of each liquid cargo pump, the processor constructs a common mode control target using the total transport volume target and uses the pressure of the common transport pipeline as the constrained state. Both methods use the same dynamic hydraulic coupling spectrum, the same common mode and differential mode decomposition mechanism, and the same propagation delay scheduling mechanism. The selection of the target variable does not change the control structure for solving the inter-cabin difference state in the differential mode subspace.

[0070] In one implementation, the inter-compartment difference state is composed of the estimated remaining unloading time of each compartment. In another implementation, the inter-compartment difference state is composed of the combination of the liquid level drop rate and the change in tank pressure. The processor normalizes the dimensions of each state quantity and uses the stable fluctuation range under the current operating conditions as the normalization scale. The normalized difference state is then used to solve the differential mode control. The unnormalized actual liquid level, tank pressure, outlet pressure, and drive current are still used to form the feasible domain boundary to avoid the direct superposition of different dimensions affecting the control direction.

[0071] Preferably, the system does not immediately enable full differential mode control during the control startup phase. The remote control processor first calls the verified spectrum closest to the current operating condition index, only obtains the common mode reference solution with limited amplitude, and uses the naturally generated speed regulation response to verify the direction of the coupling edge and the propagation delay. After the main coupling edges corresponding to the running pump set, valve connection combination and common delivery pipeline pressure range have completed the consistency verification, the differential mode control subspace and phase scheduling degrees of freedom are gradually released. If the actual pressure response is opposite to the predicted direction during the verification period, the corresponding coupling edge is frozen and the relevant control components are limited to the continuous change range allowed by the local loop.

[0072] Furthermore, during the control exit phase, the system shrinks the differential mode control degrees of freedom in the reverse order of the startup phase. The processor stops generating new cross-pump differential mode corrections, withdraws the scheduled but not yet executed differential mode sub-components to the finite prediction model, and retains only the common mode control components required to maintain a smooth transition of the current overall delivery state. After the target speed of each liquid cargo pump returns to a stable state that can be maintained independently by its respective local control loop, the remote joint control solution is stopped, so that no unpredictable multi-pump synchronous changes occur during the control switch.

[0073] In a specific operational scenario, the liquid cargo pumps corresponding to three liquid cargo tanks are simultaneously connected to a common delivery pipeline. The first and second liquid cargo pumps merge via adjacent branches, while the third liquid cargo pump merges in the downstream section. The remote control processor identifies, based on historical natural speed regulation responses, that the first and second liquid cargo pumps have similar positive gains and propagation delays to the pressure of the common delivery pipeline, while the third liquid cargo pump has a longer propagation delay. When the total delivery target requirement increases and the estimated remaining unloading time of the first liquid cargo tank is shorter than that of the other liquid cargo tanks, the processor projects the total increase as a common-mode control component shared by the three liquid cargo pumps. Within the differential-mode subspace, it reduces the component of the first liquid cargo pump and increases the components of the second and third liquid cargo pumps. Then, it arranges the same-direction pressure responses generated by the second and third liquid cargo pumps to different arrival time windows. The synthesized control quantity is then issued after verification in the feasible domain.

[0074] During the actual state feedback phase, the remote control processor calculates the prediction residual for each state variable according to the prediction time. The liquid level residual is used to verify the transport state estimation, the outlet pressure residual and the common transport pipeline pressure residual are used to verify the coupling edge gain and propagation delay, and the drive current residual is used to distinguish between pump load changes and pure pipeline pressure changes. If multiple related residuals maintain a consistent direction on the same propagation path, the residuals are assigned to the update sample of the corresponding coupling edge. If only a single state variable has an isolated residual, the current coupling parameters are maintained and the system waits for verification in subsequent control cycles. If the residual direction is continuously opposite to the prediction direction of the dynamic spectrum, the relevant coupling edge is frozen and the most recently verified parameter version under the same operating condition index is called.

[0075] In the case of multiple shared transport paths, the control changes of the same liquid pump can be transmitted to the common transport pipeline through different merging sections. The remote control processor establishes branch responses according to the propagation delay and impact gain of each path, without pre-combining all paths into a single response. During prediction, each branch response is first written into the corresponding arrival time, and then the branch responses at the same prediction time are algebraically superimposed. When the valve state changes and a branch path is disconnected, only the corresponding branch edge is removed from the current time-sharing coupling matrix, and the remaining reachable paths continue to participate in the prediction. Therefore, it is possible to retain the still valid coupling relationship after the pipeline connectivity changes and reduce the range of parameters that need to be re-identified.

[0076] When multiple coordinated control commands have the same planned execution time, the pump control execution interface caches command frames according to the scheduling time window generated by the remote control processor. Before the start of each time window, it checks the latest operating status of the corresponding liquid pump and the status of the associated valve. If the latest status is consistent with the operating condition index used in the solution, the command is sent according to the plan. If the set of operating pumps or the valve connection combination has changed, the command that has not yet been sent is canceled and a status change receipt is returned. The remote control processor reconstructs the predicted status and control sequence with the new operating condition index. Commands that have been sent but have not yet resulted in actual speed changes are retained or withdrawn according to the execution receipt to prevent the control components generated under the old operating condition from entering the hydraulic relationship under the new operating condition.

[0077] Through the above implementation methods, the dynamic hydraulic coupling map can be updated according to the participating pump sets, valve connection combinations, liquid level status, and pressure range of the common delivery pipeline. The common mode control component and the differential mode control component correspond to the overall delivery status and the inter-compartment difference status, respectively. The propagation delay scheduling arranges staggered or overlapping execution according to the expected response arrival time of each control component. Feasibility projection and marginal contribution redistribution restrict and transfer the control quantity that exceeds the feasible domain of the predicted state. The system continuously corrects the network delay, local response delay, and hydraulic propagation parameters by actual execution receipts and state responses, forming a complete control closed loop that can be implemented in the scenario of remote joint control of multi-compartment liquid cargo pumps on liquefied gas carriers.

Claims

1. A remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers, characterized in that, This includes a status acquisition interface, a remote control processor, and a pump control execution interface; The status acquisition interface acquires the liquid level and pressure of each liquid cargo tank, as well as the speed, drive current, outlet pressure, operating status, associated valve status, and common pipeline pressure of each liquid cargo pump according to a unified time base. The remote control processor constructs a dynamic hydraulic coupling spectrum that updates with the operating conditions based on the time delay response relationship and gain correlation relationship between various state variables. It establishes a multi-pump finite time domain prediction state based on the dynamic hydraulic coupling spectrum and decomposes the liquid cargo pump control quantity into a common mode control component to maintain the overall delivery target and a differential mode control component to correct the state deviation between compartments. After performing phase configuration on the differential mode control component according to the propagation delay of the coupling path, it generates a coordinated control command for each liquid cargo pump. The pump control execution interface sends the collaborative control commands to the remote control loops of the corresponding liquid cargo pumps. The remote control processor constructs a multi-pump predicted state vector based on the target remaining unloading time, liquid level drop rate, tank pressure change, and outlet pressure and drive current of each liquid cargo tank. It also calculates the mapping relationship between the control quantity of each liquid cargo pump and the overall transport state and inter-tank difference state based on the coupling gain matrix corresponding to the dynamic hydraulic coupling spectrum. The control direction that maintains the target total delivery volume or the target common delivery pipeline pressure is defined as the common mode control subspace, and the control direction that does not change the target total delivery volume and is used to adjust the differences in the predicted state of different liquid cargo tanks is defined as the differential mode control subspace. The common mode control component and the differential mode control component are obtained by orthogonal projection respectively. The remote control processor establishes a sequence of expected response arrival times for each differential control component based on the propagation delay from each liquid cargo pump to the common transport pipeline node in the dynamic hydraulic coupling diagram, and performs staggered arrangement of the same-direction pressure response and overlapping arrangement of the opposite-direction pressure response according to the expected response arrival times. Within any control cycle, the arranged differential control components and the common control components are combined to form a candidate control vector. The candidate control vector is then subjected to a feasibility projection based on the continuity of the liquid cargo pump speed, the rate of change of the outlet pressure, and the allowable pressure range of the common delivery pipeline to obtain the coordinated control command corresponding to each liquid cargo pump.

2. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 1, characterized in that, The remote control processor sets the liquid cargo tank, liquid cargo pump, and pipeline sections with shared transportation relationships as map nodes, and extracts candidate time delay sequences between any liquid cargo pump speed change, start-stop change, or associated valve status change and other map node status responses within a continuous control cycle. The candidate delay sequences are subjected to directional consistency screening and response attenuation screening to obtain coupling edges that characterize the influence direction, influence gain, and propagation delay. The coupling edges are grouped and stored according to the current set of liquid cargo pumps, valve connection status, and pressure range of the common transport pipeline. The weights of the coupling edges are updated using the response residuals of adjacent control cycles to form a dynamic hydraulic coupling map corresponding to the current transport topology and operating conditions.

3. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 2, characterized in that, The remote control processor extracts control change segments from the normal speed regulation, planned start-up and shutdown, and associated valve switching of the liquid cargo pump, and uses the stable state before the control change as the reference state to perform reference elimination and time alignment on the state increment of each graph node after the control change segment. State increments that maintain the same response direction across multiple control change segments and whose propagation delay falls into the corresponding delay cluster are used to update the coupling edge weights. State increments that appear only in a single control change segment or are inconsistent with the pressure change direction of the common delivery pipeline are removed from the edge weight update data. When the gain change rate of the same coupling side remains in the same direction in the continuous control change segment and the propagation delay migrates from the original delay cluster to another delay cluster, a new operating condition group is established and the coupling side parameters of the original operating condition group are retained.

4. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 3, characterized in that, The remote control processor calculates the total transport state trajectory corresponding to the common mode control component and the inter-tank deviation state trajectory corresponding to the differential mode control component within the finite prediction time domain, and converts the minimum liquid level constraint, liquid level drop rate constraint, drive current constraint, outlet pressure constraint and control quantity change rate constraint of each liquid cargo tank into the prediction state feasible domain. First, a common-mode reference solution that satisfies the total transport target is obtained within the feasible region of the predicted state. Then, a differential-mode correction solution is obtained in the null space that keeps the total transport trajectory corresponding to the common-mode reference solution unchanged. The candidate speed adjustment amount of each liquid cargo pump is formed according to the sum of the common-mode reference solution and the differential-mode correction solution.

5. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 4, characterized in that, The remote control processor maps the expected arrival time of each differential control component to a continuous scheduling time window, and divides the differential control components into in-phase superposition group, out-of-phase cancellation group and weak coupling group according to the influence direction and influence gain of the coupling side. A mutually exclusive arrival time window is set for the differential mode control components in the in-phase superposition group, an overlapping arrival time window is set for the differential mode control components in the out-of-phase cancellation group, and the original control period is retained for the differential mode control components in the weakly coupled group. When any control component exceeds the speed change rate constraint of the corresponding liquid cargo pump after adjustment by the time window, the excess part will be deferred to the adjacent time window and the subsequent prediction state will be corrected synchronously.

6. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 5, characterized in that, The remote control processor calculates the marginal contribution vector of each liquid cargo pump unit control increment to the peak pressure of the common transport pipeline, the deviation of the target total transport volume, and the deviation of the predicted unloading time between compartments, based on the dynamic hydraulic coupling map, for the candidate control components that are reduced in the feasibility projection. A redistribution priority sequence is constructed based on the remaining liquid level, outlet pressure margin, drive current margin, and marginal contribution vector of the corresponding coupling edge of each liquid cargo pump. Under the condition of keeping the predicted pressure trajectory of the common delivery pipeline within the allowable range, the reduced delivery control quantity is gradually allocated to other liquid cargo pumps in the priority sequence to form a cooperative control command constrained by marginal contribution.

7. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 6, characterized in that, After establishing a new operating condition group, the remote control processor encodes the set of pumps, valve connection combinations, pressure range of common delivery pipeline, and liquid level range participating in the current control cycle into an operating condition index, and selects the coupling edge parameter with the smallest distance from the operating condition index from the original operating condition group as the initial parameter. Before the consistency verification of the response samples that simultaneously contain the liquid cargo pump speed change segment and the associated valve switching segment is completed, a recursive confidence weight is set for the initial parameters, and the newly collected response and the original operating condition prediction response are fused according to the recursive confidence weight. After the consistency verification is completed, only the coupling edges that have passed the consistency verification of direction and propagation delay are updated, while the coupling edges that have not obtained valid response samples are frozen, forming a dynamic hydraulic coupling map constrained by confidence migration.

8. The remote intelligent control system for multi-compartment cargo pumps on liquefied gas carriers according to claim 7, characterized in that, When the remote control processor performs the redistribution of the transport control quantity, it divides the liquid cargo pumps in the priority sequence into a main receiving set and a compensation receiving set according to the predicted suction safety margin and the predicted remaining unloading time. First, find the first redistribution vector within the main receiving set that minimizes the peak pressure increment of the common delivery pipeline. Then, project the delivery control quantities not received by the first redistribution vector onto the feasible control subspace of the compensation receiving set to obtain the second redistribution vector. Propagation delay correction is performed on the first redistribution vector and the second redistribution vector respectively, and the order of the main receiving set and the compensation receiving set is adjusted in subsequent control cycles according to the deviation between the actual pressure response and the predicted pressure response.

Images